The fundamental problem is that natural language processing (NLP) isn't one problem to be solved. There will be many solutions to many particular problems of NLP. Many, many years later we're still waiting for one solution to energy (steam, or coal, or nuclear...?) or one solution to flight (gliders, or jets, or rockets...?). Human intelligence is a swiss army knife where the tools developed millions of years ago (opposable thumbs) or evolved withing ten years (keyword-based search). Why should there be a general solution?

Google has taken a sizeable chunk. But what's interesting to me is search has become all-consuming for them. Many solutions are perfectly fine for search, but that perspective (aggregating large noisy datasets) won't help much in other areas (e.g., individual voice recognition). If anything it will be a big waste of time trying to shoehorn every NLP problem into that search-based rubric. That leaves a lot of room for little guys cranking through data to produce new, and unique, NLP tools (e.g., Dragon Naturally Speaking). More to your point: I have yet to see a scientific hire by Facebook away from Google. It's been all business and marketing from what I've seen in the popular press.

"Why should there be a general solution?"

Why indeed. However, finding one was very much on the agenda for GOFAI, and remains on the agenda for various AGI efforts, e.g. by people like Pei Wang, Ben Goertzel, and Eliezer Yudkowsky (who comments here, as I've observed).

> individual voice recognition

That's where 1-800-GOOG-411 comes in.